Tutorial for sorting data stored as numpy to on-resonance R1rho analysis

Data background

This is data recorded at 600 and 950 MHz.
This should follow relax-4.0.1.Darwin.dmg installation on a Mac, only with GUI.

For each spectrometer frequency, the data is saved in np.arrays

1. one for the residue number,
2. one for the rates,
3. one for the errorbars,
4. one for the RF field strength.

They can be retrieved also with scipy's loadmat command.

The experiments are on-resonance R_1ρ, and the rates are already corrected for the (small) offset effect, using the experimentally determined R₁.

Specifically, the numpy shapes of the data is:

1. For 600 MHz
   1. residues (1, 60)
   2. rates (60, 10)
   3. errorbars_rate (60, 10)
   4. RFfields (1, 10)

1. For 950 Mhz
   1. residues (1, 61)
   2. rates (61, 19)
   3. errorbars_rate (61, 19)
   4. RFfields (1, 19)

For the 950 MHz, the 4000 Hz dispersion point is replicated.

An example of the data at the 2 fields is:

• 

  600 MHz

• 

  950 MHz

Create data files for relax

First prepare data, by running in Python:

python 1_prepare_data.py

The script is:

Python 2/3 script: The 1_prepare_data.py script.

import os
from scipy.io import loadmat
import numpy as np

# Set path
#cwd = status.install_path+os.sep+'test_suite'+os.sep+'shared_data'+os.sep+'dispersion'+os.sep+'Paul_Schanda_2015_Nov'
cwd = os.getcwd()
outdir = cwd + os.sep

fields = [600, 950]
file_names = ['residues', 'rates', 'errorbars_rate', 'RFfields']

# Store data in dictionary
all_data = {}
all_data['fields'] = fields
all_data['file_names'] = file_names

# Make list of residues and make unique
all_res = []

# Loop over the experiments, collect all data
for field in fields:
    print("%s"%field,)

    # Make a dic inside
    all_data['%s'%field] = {}

    # Construct the path to the data
    path = cwd + os.sep

    # Collect all filename paths
    field_file_name_paths = []
    for file_name in file_names:
        # Create path name
        file_name_path = path + "%s_%s.mat"%(field, file_name)
        field_file_name_paths.append(file_name_path)

        # Load the data
        file_name_path_data = loadmat(file_name_path)
        # Extract as numpy
        file_name_path_data_np = file_name_path_data[file_name]
        # And store
        all_data['%s'%field]['%s'%file_name] = file_name_path_data
        all_data['%s'%field]['np_%s'%file_name] = file_name_path_data_np

        print(file_name, file_name_path_data_np.shape)

        # Collect residues
        if file_name == "residues":
            all_res += list(file_name_path_data_np.flatten())

    # Store
    all_data['%s'%field]['field_file_name_paths'] = field_file_name_paths


# Make list of residues and make unique
all_res_uniq = sorted(list(set(all_res)))
all_data['all_res_uniq'] = all_res_uniq

# Write a sequence file for relax
f = open(outdir + "residues.txt", "w")
f.write("# Residue_i\n")
for res in all_res_uniq:
    f.write("%s\n"%res)
f.close()

f_exp = open(outdir + "exp_settings.txt", "w")
f_exp.write("# sfrq_MHz RFfield_kHz file_name\n")

# Then write the files for the rates
k = 1
for field in all_data['fields']:
    resis = all_data['%s'%field]['np_residues'][0]
    rates = all_data['%s'%field]['np_rates']
    errorbars_rate = all_data['%s'%field]['np_errorbars_rate']
    RFfields = all_data['%s'%field]['np_RFfields'][0]

    print("\nfield: %3.3f"%field)
    for i, RF_field_strength_kHz in enumerate(RFfields):
        #print "RF_field_strength_kHz: %3.3f"%RF_field_strength_kHz
        # Generate file name
        f_name = outdir + "sfrq_%i_MHz_RFfield_%1.3f_kHz_%03d.in"%(field, RF_field_strength_kHz, k)
        cur_file = open(f_name, "w")
        cur_file.write("# resi rate        rate_err\n")

        exp_string = "%11.7f %11.7f %s\n"%(field, RF_field_strength_kHz, f_name)
        print("%s"%exp_string,)
        f_exp.write(exp_string)

        for j, resi in enumerate(resis):
            rate = rates[j, i]
            error = errorbars_rate[j, i]
            string = "%4d %11.7f %11.7f\n"%(resi, rate, error)
            cur_file.write(string)

        cur_file.close()
        k += 1

f_exp.close()

Run analysis in relax GUI

• Start relax
• Then click: new analysis or ⌘ Cmd+n
• Then click icon for Relaxation dispersion → Next
• Just accept name for the pipe

• Open the interpreter: ⌘ Cmd+p
• Paste in

import os; os.chdir(os.getenv('HOME') + os.sep + 'Desktop' + os.sep + 'temp'); pwd()

• Then do

script(file='2_load_data.py')

• ( You can scroll through earlier commands with: ⌘ Cmd+↑ )
• Close the Grace Results viewer window
• Close the interpreter window

Now save the current state! This saved state can now be loaded just before an analysis, and contain all setup and data.

• File  → Save as (Shift+Ctrl+S) → Save as before_analysis.bz2

Now Quit relax, and start relax again.
Go to: File  → Open relax state (Ctrl+O) → Locate before_analysis.bz2

In the GUI, select:

• For models, only select: No Rex and M61
• R₁ parameter optimisation to False, as this on-resonance data does not use R₁.
• The number of Monte-Carlo simulations to 3. (The minimum number before relax refuse to run).
• Let other things be standard, and click execute

Note  The number of Monte-Carlo simulations is set to 3. This means that relax will do the analysis, and the fitted parameters will be correct, but the standard error of the parameters will be wrong.

The reason for this is, that the data "naturally"? for some spins contains measurements which are "doubtful".
When relax is performing Monte-Carlo simulations, all the data is first copied x times to the number of Monte-Carlo simulations, and each R_2,eff point on the graph is modified randomly by a gaussian noise with a width described by the associated errors.

For spins which contains measurements which are "doubtful", relax will be spending "very long time" trying to fit a meaningful model.
This "time" is poorly spend on "bad data".

Rather, one should first try to

• analyse the data quickly
• make the graphs
• examine which spins should be deselected
• and then rerun the analysis with a higher number of monte-carlo simulations

After this comes the part with global fit/clustered analysis.

• analyse the data again
• Select which spins should be included in a global fit/clustered analysis
• and then rerun with analysis

The script

relax script: The 2_load_data.py script.

###############################################################################
#                                                                             #
# Copyright (C) 2015 Troels E. Linnet                                         #
#                                                                             #
# This file is part of the program relax (http://www.nmr-relax.com).          #
#                                                                             #
# This program is free software: you can redistribute it and/or modify        #
# it under the terms of the GNU General Public License as published by        #
# the Free Software Foundation, either version 3 of the License, or           #
# (at your option) any later version.                                         #
#                                                                             #
# This program is distributed in the hope that it will be useful,             #
# but WITHOUT ANY WARRANTY; without even the implied warranty of              #
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
# GNU General Public License for more details.                                #
#                                                                             #
# You should have received a copy of the GNU General Public License           #
# along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
#                                                                             #
###############################################################################
# Test if running as script or through GUI.
is_script = False
if not hasattr(cdp, "pipe_type"):
    is_script = True
    # We need to create a data pipe, which will tell relax which type of data we are expecting 
    pipe_name = 'relax_disp'
    pipe_bundle = 'relax_disp'
    pipe.create(pipe_name, pipe_bundle)

# Minimum: Just read the sequence data, but this misses a lot of information.
sequence.read(file='residues.txt', res_num_col=1)

# Open the settings file
set_file = open("exp_settings.txt")
set_file_lines = set_file.readlines()

for line in set_file_lines:
    if "#" in line[0]:
        continue

    # Get data
    field, RF_field_strength_kHz, f_name = line.split()

    # Assign data
    spec_id = f_name
    relax_disp.exp_type(spectrum_id=spec_id, exp_type='R1rho')

    # Set the spectrometer frequency
    spectrometer.frequency(id=spec_id, frq=float(field), units='MHz')

    # Is in kHz, som convert to Hz
    #http://wiki.nmr-relax.com/Relax_disp.spin_lock_offset%2Bfield
    #http://www.nmr-relax.com/manual/relax_disp_spin_lock_field.html
    disp_frq = float(RF_field_strength_kHz)*1000

    # Set The spin-lock field strength, nu1, in Hz
    relax_disp.spin_lock_field(spectrum_id=spec_id, field=disp_frq)

    # Read the R2eff data
    relax_disp.r2eff_read(id=spec_id, file=f_name, dir=None, disp_frq=disp_frq, res_num_col=1, data_col=2, error_col=3)

    # Is this necessary? The time, in seconds, of the relaxation period.
    #relax_disp.relax_time(spectrum_id=spec_id, time=time_sl)


# Name the isotope for field strength scaling.
spin.isotope(isotope='15N')
relax_disp.select_model(model='R2eff')

# Plot data
relax_disp.plot_disp_curves(dir='grace', y_axis='r2_eff', x_axis='disp', num_points=1000, extend_hz=500.0, extend_ppm=500.0, interpolate='disp', force=True)

state.save("temp_state", force=True)

# Do it through script
if False:
#if True:
#if is_script:
    import os
    from auto_analyses import relax_disp as aa_relax_disp
    from lib.dispersion.variables import EXP_TYPE_CPMG_DQ, EXP_TYPE_CPMG_MQ, EXP_TYPE_CPMG_PROTON_MQ, EXP_TYPE_CPMG_PROTON_SQ, EXP_TYPE_CPMG_SQ, EXP_TYPE_CPMG_ZQ, EXP_TYPE_LIST, EXP_TYPE_R1RHO, MODEL_B14_FULL, MODEL_CR72, MODEL_CR72_FULL, MODEL_DPL94, MODEL_IT99, MODEL_LIST_ANALYTIC_CPMG, MODEL_LIST_FULL, MODEL_LIST_NUMERIC_CPMG, MODEL_LM63, MODEL_M61, MODEL_M61B, MODEL_MP05, MODEL_NOREX, MODEL_NS_CPMG_2SITE_3D_FULL, MODEL_NS_CPMG_2SITE_EXPANDED, MODEL_NS_CPMG_2SITE_STAR_FULL, MODEL_NS_R1RHO_2SITE, MODEL_NS_R1RHO_3SITE, MODEL_NS_R1RHO_3SITE_LINEAR, MODEL_PARAMS, MODEL_R2EFF, MODEL_TP02, MODEL_TAP03
    # Number of grid search increments.  If set to None, then the grid search will be turned off and the default parameter values will be used instead.
    #GRID_INC = None
    GRID_INC = 21
    # The number of Monte Carlo simulations to be used for error analysis at the end of the analysis.
    MC_NUM = 3
    # Model selection technique.
    MODSEL = 'AIC'
    result_dir_name = os.getcwd() + os.sep + "results"
    # Which models to analyse ?
    #MODELS = [MODEL_R2EFF, MODEL_NOREX, MODEL_DPL94, MODEL_TP02, MODEL_TAP03, MODEL_MP05, MODEL_NS_R1RHO_2SITE]
    #MODELS = [MODEL_R2EFF, MODEL_NOREX]
    # R2EFF shall be skipped here
    #MODELS = [MODEL_NOREX]
    # http://wiki.nmr-relax.com/M61
    MODELS = [MODEL_NOREX, MODEL_M61]

    # Fit, instead of read
    #r1_fit = True
    r1_fit = False

    # Set the initial guess from the minimum R2eff point
    set_grid_r20=True

    """
        @keyword pipe_name:                 The name of the data pipe containing all of the data for the analysis.
        @type pipe_name:                    str
        @keyword pipe_bundle:               The data pipe bundle to associate all spawned data pipes with.
        @type pipe_bundle:                  str
        @keyword results_dir:               The directory where results files are saved.
        @type results_dir:                  str
        @keyword models:                    The list of relaxation dispersion models to optimise.
        @type models:                       list of str
        @keyword grid_inc:                  Number of grid search increments.  If set to None, then the grid search will be turned off and the default parameter values will be used instead.
        @type grid_inc:                     int or None
        @keyword mc_sim_num:                The number of Monte Carlo simulations to be used for error analysis at the end of the analysis.
        @type mc_sim_num:                   int
        @keyword exp_mc_sim_num:            The number of Monte Carlo simulations for the error analysis in the 'R2eff' model when exponential curves are fitted.  This defaults to the value of the mc_sim_num argument when not given.  When set to '-1', the R2eff errors are estimated from the Covariance matrix.  For the 2-point fixed-time calculation for the 'R2eff' model, this argument is ignored.
        @type exp_mc_sim_num:               int or None
        @keyword modsel:                    The model selection technique to use in the analysis to determine which model is the best for each spin cluster.  This can currently be one of 'AIC', 'AICc', and 'BIC'.
        @type modsel:                       str
        @keyword pre_run_dir:               The optional directory containing the dispersion auto-analysis results from a previous run.  The optimised parameters from these previous results will be used as the starting point for optimisation rather than performing a grid search.  This is essential for when large spin clusters are specified, as a grid search becomes prohibitively expensive with clusters of three or more spins.  At some point a RelaxError will occur because the grid search is impossibly large.  For the cluster specific parameters, i.e. the populations of the states and the exchange parameters, an average value will be used as the starting point.  For all other parameters, the R20 values for each spin and magnetic field, as well as the parameters related to the chemical shift difference dw, the optimised values of the previous run will be directly copied.
        @type pre_run_dir:                  None or str
        @keyword optimise_r2eff:            Flag to specify if the read previous R2eff results should be optimised.  For R1rho models where the error of R2eff values are determined by Monte-Carlo simulations, it can be valuable to make an initial R2eff run with a high number of Monte-Carlo simulations.  Any subsequent model analysis can then be based on these R2eff values, without optimising the R2eff values.
        @type optimise_r2eff:               bool
        @keyword insignificance:            The R2eff/R1rho value in rad/s by which to judge insignificance.  If the maximum difference between two points on all dispersion curves for a spin is less than this value, that spin will be deselected.  This does not affect the 'No Rex' model.  Set this value to 0.0 to use all data.  The value will be passed on to the relax_disp.insignificance user function.
        @type insignificance:               float
        @keyword numeric_only:              The class of models to use in the model selection.  The default of False allows all dispersion models to be used in the analysis (no exchange, the analytic models and the numeric models).  The value of True will activate a pure numeric solution - the analytic models will be optimised, as they are very useful for replacing the grid search for the numeric models, but the final model selection will not include them.
        @type numeric_only:                 bool
        @keyword mc_sim_all_models:         A flag which if True will cause Monte Carlo simulations to be performed for each individual model.  Otherwise Monte Carlo simulations will be reserved for the final model.
        @type mc_sim_all_models:            bool
        @keyword eliminate:                 A flag which if True will enable the elimination of failed models and failed Monte Carlo simulations through the eliminate user function.
        @type eliminate:                    bool
        @keyword set_grid_r20:              A flag which if True will set the grid R20 values from the minimum R2eff values through the r20_from_min_r2eff user function. This will speed up the grid search with a factor GRID_INC^(Nr_spec_freq). For a CPMG experiment with two fields and standard GRID_INC=21, the speed-up is a factor 441.
        @type set_grid_r20:                 bool
        @keyword r1_fit:                    A flag which if True will activate R1 parameter fitting via relax_disp.r1_fit for the models that support it.  If False, then the relax_disp.r1_fit user function will not be called.
    """


    # Go
    aa_relax_disp.Relax_disp(pipe_name=pipe_name, pipe_bundle=pipe_bundle, results_dir=result_dir_name, models=MODELS, grid_inc=GRID_INC, mc_sim_num=MC_NUM, exp_mc_sim_num=None, modsel=MODSEL,  pre_run_dir=None, optimise_r2eff=False, insignificance=0.0, numeric_only=False, mc_sim_all_models=False, eliminate=True, set_grid_r20=set_grid_r20, r1_fit=r1_fit)

Run the same analysis in relax through terminal

The analysis can be performed by

python 1_prepare_data.py

relax 2_load_data.py -t log.txt
# Or
mpirun -np 8 relax --multi='mpi4py' 2_load_data.py -t log.txt

Inspect and make graphs

• When this is done, quit relax
• Then to convert all xmgrace files to png

First make graphs

cd grace
./grace2images.py -t EPS,PNG

cd ..
cd No_Rex/
./grace2images.py -t EPS,PNG
cat r1rho_prime.out

cd ..
cd M61
./grace2images.py -t PNG
cat r1rho_prime.out
cat kex.out
cat phi_ex.out 

cat kex.out | grep "None          12"
cat phi_ex.out | grep "None          12"

This should give the images of the data

• 

  Resi 12 at 600+950 MHz

For model M61, spin 12 is:

• 

  Resi 12 at 600+950 MHz for M61 model. kex=25000, phi_ex=0.32

Further analysis steps

Load the final state

• Inspect which residues should be included in analysis

# Now simulate that all spins are first deselected, and then selected.
deselect.all()
sel_ids = [
":12",
":51",
]
for sel_spin in sel_ids:
    print("Selecting spin %s"%sel_spin)
    select.spin(spin_id=sel_spin, change_all=False)

# Inspect which residues should be analysed together for a clustered/global fit.
cluster_ids = sel_ids

# Cluster spins
for curspin in cluster_ids:
    print("Adding spin %s to cluster"%curspin)
    relax_disp.cluster('model_cluster', curspin)

# Show the pipe
print("\nPrinting all the available pipes.")
pipe.display()

# Get the selected models
print("\nChecking which model is stored per spin.")
for curspin, mol_name, res_num, res_name, spin_id in spin_loop(full_info=True, return_id=True, skip_desel=False):
    print("For spin_id '%s the model is '%s''"%(spin_id, curspin.model))

# Copy pipe and switch to it.
pipe.copy(pipe_from="final - relax_disp", pipe_to="relax_disp_cluster", bundle_to="relax_disp_cluster")
pipe.switch(pipe_name="relax_disp_cluster")
pipe.display()

# Go again with clustered spins.
relax_disp.Relax_disp(pipe_name="relax_disp_cluster", pipe_bundle="relax_disp_cluster", results_dir=outdir+sep+"cluster", models=MODELS, grid_inc=GRID_INC, mc_sim_num=MC_NUM, exp_mc_sim_num=None, modsel=MODSEL,  pre_run_dir=None, optimise_r2eff=False, insignificance=0.0, numeric_only=False, mc_sim_all_models=False, eliminate=True, set_grid_r20=set_grid_r20, r1_fit=r1_fit)

# Get the clustered fitted values
print("\nChecking which value is stored per spin.")
kex = None
for curspin, mol_name, res_num, res_name, spin_id in spin_loop(full_info=True, return_id=True, skip_desel=False):
    if kex == None:
        kex = curspin.kex
    self.assertEqual(curspin.kex, kex)
    print("For spin_id %s the kex is %.3f"%(spin_id, kex))

• Run analysis again